Update src/populate.py

src/populate.py

@@ -8,32 +8,78 @@ from src.display.utils import AutoEvalColumn, EvalQueueColumn
 from src.leaderboard.read_evals import get_raw_eval_results
 
 
+# def get_leaderboard_df(results_path: str, requests_path: str, cols: list, benchmark_cols: list) -> pd.DataFrame:
+#     """Creates a dataframe from all the individual experiment results"""
+#     raw_data = get_raw_eval_results(results_path, requests_path)
+#     all_data_json = [v.to_dict() for v in raw_data]
+
+
+
+#     df = pd.DataFrame.from_records(all_data_json)
+#     score_cols = [
+#         'ALT E to J BLEU', 'ALT J to E BLEU', 'WikiCorpus E to J BLEU', 'WikiCorpus J to E BLEU', 
+#         'XL-Sum JA BLEU', 'XL-Sum ROUGE1', 'XL-Sum ROUGE2', 'XL-Sum ROUGE-Lsum'
+#     ]
+    
+#     existing_score_cols = [col for col in score_cols if col in df.columns]
+#     print(f"Existing score columns: {existing_score_cols}")
+    
+#     # スコア列を100で割り、.4f形式でフォーマット
+#     df[existing_score_cols] = (df[existing_score_cols] / 100).applymap(lambda x: f'{x:.4f}')
+#     df = df.sort_values(by=[AutoEvalColumn.AVG.name], ascending=False)
+#     df = df[cols].round(decimals=2)
+    
+#     # filter out if any of the benchmarks have not been produced
+#     df = df[has_no_nan_values(df, benchmark_cols)]
+#     return df
+
 def get_leaderboard_df(results_path: str, requests_path: str, cols: list, benchmark_cols: list) -> pd.DataFrame:
     """Creates a dataframe from all the individual experiment results"""
     raw_data = get_raw_eval_results(results_path, requests_path)
+    
+    # デバッグ: Raw data の内容を確認
+    print(f"Raw data before conversion: {raw_data}")
+    
     all_data_json = [v.to_dict() for v in raw_data]
 
-
-
+    # デバッグ: 全てのデータをDataFrameに変換
     df = pd.DataFrame.from_records(all_data_json)
+    print(f"Initial DataFrame: {df}")
+    
     score_cols = [
         'ALT E to J BLEU', 'ALT J to E BLEU', 'WikiCorpus E to J BLEU', 'WikiCorpus J to E BLEU', 
         'XL-Sum JA BLEU', 'XL-Sum ROUGE1', 'XL-Sum ROUGE2', 'XL-Sum ROUGE-Lsum'
     ]
     
+    # デバッグ: 存在するスコア列を確認
     existing_score_cols = [col for col in score_cols if col in df.columns]
     print(f"Existing score columns: {existing_score_cols}")
     
     # スコア列を100で割り、.4f形式でフォーマット
     df[existing_score_cols] = (df[existing_score_cols] / 100).applymap(lambda x: f'{x:.4f}')
+    
+    # デバッグ: スコア調整後のデータフレームを確認
+    print(f"DataFrame after score adjustment: {df}")
+    
+    # ソート
     df = df.sort_values(by=[AutoEvalColumn.AVG.name], ascending=False)
-    df = df[cols].round(decimals=2)
     
-    # filter out if any of the benchmarks have not been produced
+    # デバッグ: ソート後のデータフレームを確認
+    print(f"Sorted DataFrame: {df}")
+    
+    # NaNを持つ行を除外
     df = df[has_no_nan_values(df, benchmark_cols)]
+    
+    # デバッグ: NaNフィルタリング後のデータフレームを確認
+    print(f"Final DataFrame after NaN filtering: {df}")
+    
+    # 必要なカラムのみに絞り込む
+    df = df[cols].round(decimals=2)
+    
     return df
 
 
+
 def get_evaluation_queue_df(save_path: str, cols: list) -> list[pd.DataFrame]:
     """Creates the different dataframes for the evaluation queues requestes"""
     entries = [entry for entry in os.listdir(save_path) if not entry.startswith(".")]